HG-Bitmap Join Index: A Hybrid GPU/CPU Bitmap Join Index Mechanism for OLAP

Abstract

In-memory big data OLAP(on-line analytical processing) is time consuming task for data access latency and complex star join processing overhead. GPU is introduced to DBMSs for its remarkable parallel computing power but also restricted by its limited GPU memory size and low PCI-E bandwidth between GPU and memory. GPU is suitable for linear processing with its powerful SIMD(Single Instruction Multiple Data) parallel processing, and lack efficiency for complex control and logic processing. So how to optimize management for dimension tables and fact table, how to dispatch different processing stages of OLAP(Select, Project, Join, Grouping, Aggregate) between CPU and GPU devices and how to minimize data movement latency and maximize parallel processing efficiency of GPU are important for a hybrid GPU/CPU OLAP platform. We propose a hybrid GPU/CPU Bitmap Join index(HG-Bitmap Join index) for OLAP to exploit a GPU memory resident join index mechanism to accelerate star join in a star schema OLAP workload. We design memory constraint bitmap join index with fine granularity keyword based bitmaps from TOP K predicates to accurately assign specified GPU memory size for specified frequent keyword bitmap join indexes. An OLAP query is transformed into bitwise operations on matched bitmaps first to generate global bitmap filter to minimize big fact table scan cost. In this mechanism, GPU is fully utilized with simple bitmap store and processing, the small bitmap filter from GPU to memory minimizes the data movement overhead, and the hybrid GPU/CPU join index can improve OLAP performance dramatically.